Bayesian dynamic linear models for structural health monitoring

Struct Control Health Monit

2018

Self Cite

Summary The detection of changes in structural behaviour over time, that is, anomalies, is an important aspect in structural safety analysis. This paper proposes an anomaly detection method that combines the existing Bayesian Dynamic Linear Models framework with the Switching Kalman Filter theory. The key aspect of this method is its capacity to detect anomalies based on the prior probability of an anomaly, a generic anomaly model, as well as transition probabilities between a normal and an abnormal state. Moreover, the approach operates in a semisupervised setup where normal and abnormal state labels are not required to train the model. The potential of the new method is illustrated on the displacement data recorded on a dam in Canada. The results show that the approach succeeded in identifying the anomaly caused by refection work, without triggering any false alarm. It also provided the specific information about the dam's health and conditions.

Section: Switching Kalman Filtermentioning

confidence: 99%

Anomaly detection with the Switching Kalman Filter for structural health monitoring

Nguyen

Struct Control Health Monit

2018

Self Cite

“…and Koo, K. (2017). The specific mathematical formulation associated with each component is detailed either by West and Harisson (1999) and by Goulet (2017). Components should be seen as building blocks which when assembled together, are able to model a wide variety of behaviour.…”

Section: Bayesian Dynamic Linear Modelsmentioning

confidence: 99%

“…Block matrices with sine and cosine components are defined using the generic formulation for periodic components presented by Goulet (2017) and West and Harisson (1999). The model error covariance is…”

Section: Temperature -Y T Tmentioning

confidence: 99%

“…Optimal values P * for unknown parameters are estimated using the parameter-wise EM algorithm because the matrix-wise approach was repeatedly caught in local maxima (Goulet 2017). The parameters values maximizing the log-likelihood are: …”

Section: Model Parameter Estimationmentioning

confidence: 99%

“…More recently, Goulet (2017) has adapted the BDLM theory for the specificities of the SHM context. This perform an empirical validation of BDLM in the context of Structural Health Monitoring (SHM), for separating the observed response of a structure into sub-components.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Empirical Validation of Bayesian Dynamic Linear Models in the Context of Structural Health Monitoring

Koo

2018

J. Bridge Eng.

Bayesian Dynamic Linear Models (BDLM) are traditionally employed in the fields of applied statistics and Machine Learning. This paper performs an empirical validation of BDLM in the context of Structural Health Monitoring (SHM) for separating the observed response of a structure into subcomponents. These sub-components describe the baseline response of the structure, the effect of traffic, and the effect of temperature. This utilization of BDLM for SHM is validated with data recorded on the Tamar Bridge (UK). This study is performed in the context of large-scale civil structures where missing data, outliers and non-uniform time steps are present. The study shows that the BDLM is able to separate observations into generic sub-components allowing to isolate the baseline behavior of the structure.

Uncertainty quantification for model parameters and hidden state variables in Bayesian dynamic linear models

Nguyen

Gaudot

Struct Control Health Monit

2018

Self Cite

Summary The quantification of uncertainty associated with the model parameters and the hidden state variables is a key missing aspect for the existing Bayesian dynamic linear models. This paper proposes two methods for carrying out the uncertainty quantification task: (a) the maximum a posteriori with the Laplace approximation procedure (LAP‐P) and (b) the Hamiltonian Monte Carlo procedure (HMC‐P). A comparative study of LAP‐P with HMC‐P is conducted on simulated data as well as real data collected on a dam in Canada. The results show that the LAP‐P is capable to provide a reasonable estimation without requiring a high computation cost, yet it is prone to be trapped in local maxima. The HMC‐P yields a more reliable estimation than LAP‐P, but it is computationally demanding. The estimation results obtained from both LAP‐P and HMC‐P tend to the same values as the size of the training data increases. Therefore, a deployment of both LAP‐P and HMC‐P is suggested for ensuring an efficient and reliable estimation. LAP‐P should first be employed for the model development and HMC‐P should then be used to verify the estimation obtained using LAP‐P.